Mechanical Objectivity

In the AI Story

The paradox that Daston and Galison identified is that mechanical objectivity concealed human judgment rather than eliminating it. The photograph encoded interpretive choices at every stage: framing, angle, magnification, light, lens, exposure, chemistry, paper. Each choice shaped what the image revealed and suppressed. But these choices were embedded in the apparatus rather than visible in the marks of a hand, and the users' evaluative competencies — developed through centuries of experience with manually produced images — were not equipped to recognize interpretation when it took the form of exposure settings and chemical sensitivity.

The regime produced its own characteristic scandals when the apparent objectivity of photographic evidence was revealed to be mediated in ways users had not anticipated. In astronomy, features recorded on photographic plates proved to be artifacts of the emulsion. In forensics, images presented as direct records of crime scenes carried narrative implications produced by framing choices. In anthropology, photographs presented as objective documents of cultural practices reflected the photographer's preconceptions. Each scandal motivated institutional reform — authentication protocols, technical documentation standards, training programs in photographic interpretation — but each reform addressed specific failure modes rather than the underlying structural problem.

The deeper problem was that mechanical objectivity had confused absence of visible judgment with absence of judgment. Because the interpretive work was done by the apparatus rather than by a hand the viewer could see, users treated the image as judgment-free when it was merely judgment-hidden. The confidence artifact — the sharpness, detail, and mechanical provenance of the image — inspired trust that the process did not warrant, and the gap between suggested reliability and actual reliability took decades to narrow through the slow development of critical photographic literacy.

The direct relevance to AI is the structural inheritance: today's systems are defended using arguments that would have been familiar to mid-nineteenth-century photographers. The machine does not interpret, it records. The machine does not impose theoretical commitments; it processes data. The machine produces outputs through a causal chain independent of any individual human judgment. Each of these claims has the same form as mechanical objectivity's founding claims, and each is partially false in the same way. Interpretation has been embedded in the apparatus — in training data, architectures, reward functions — and the users' existing evaluative competencies are not equipped to see interpretation when it takes these forms.

Origin

Daston and Galison located the emergence of mechanical objectivity in the specific institutional context of mid-nineteenth-century European science, when the authority of traditional scientific elites was being challenged and new instruments (the photograph, the self-registering measurement device, the automatic recorder) offered an apparent escape from the problem of individual judgment. The new ideal was simultaneously an epistemological doctrine and a moral commitment: the scientist was required to suppress herself — her preferences, her theoretical enthusiasms, her aesthetic sensibilities — in service of a representation that would be as independent as possible of her particular existence.

The moral rigor of this demand, Daston emphasized, should not be underestimated. Mechanical objectivity was not a lowering of standards but a tightening. It required a form of ascetic self-discipline in which the observer's own insight and expertise became epistemic liabilities rather than assets. That this demand produced its own characteristic distortions does not diminish the seriousness of the attempt; it only reveals that no epistemic ideal eliminates distortion, and each ideal's characteristic failures become legible only from the vantage of the regime that succeeds it.

Key Ideas

Suppression of the self as epistemic ideal. Reliable knowledge requires removing the observer's judgment from the production process, achieved through instruments that record causally rather than interpretively.

Causation as warrant of trust. The photograph's authority derived not from the photographer's expertise but from the physical chain running from specimen through lens to emulsion, supposedly independent of any consciousness.

Interpretation migrated rather than eliminated. Judgment moved from the visible hand into the apparatus — framing, optics, chemistry, development — where users lacked the competencies to detect it.

Sharpness and detail as confidence artifacts. The surface properties that photography uniquely afforded became proxies for accuracy, with the correlation strong enough to sustain trust and imperfect enough to produce systematic misreading.

The blueprint for AI's authority claims. Contemporary defenses of algorithmic objectivity reproduce nineteenth-century arguments about photographic objectivity, with interpretation once again embedded in an apparatus users cannot readily inspect.

Debates & Critiques

A recurrent debate concerns whether mechanical objectivity was ever sincerely held by working scientists or whether it functioned primarily as a public rhetoric that insiders knew to be qualified. Defenders of the concept point to extensive primary-source evidence of scientists training themselves in the ascetic self-discipline the regime required; skeptics point to equally extensive evidence that practitioners quietly relied on their expert judgment even while publicly disavowing it. The likely truth is that mechanical objectivity was both a genuine epistemic commitment and a rhetorical resource, with the tension between the two producing the characteristic pathologies Daston's framework diagnoses.